What an ATS actually does (and what it doesn't)

“ATS” has become the boogeyman of job hunting: a robot that reads your resume, scores it out of a hundred, and bins you below some threshold. That picture is mostly wrong, and believing it leads to genuinely bad decisions — white text stuffed with keywords, skills sections listing things you have never touched, and enormous anxiety about a system that is far more boring than its reputation.

What it is

An applicant tracking system is the software a company uses to manage hiring. Greenhouse, Lever, Workday, Ashby and a long tail of others. Its main job is administrative: collect applications in one place, move candidates through stages, schedule interviews, record feedback, and keep the paperwork that employment law requires.

When your resume arrives, it does three things. It stores the file. It extracts the text so the file is searchable. And it puts a row in a list for a recruiter to work through.

What it mostly does not do

It usually does not auto-reject you

Most systems can be configured to filter on knockout questions — work authorisation, a required certification, years of experience — and those are usually questions you answered on the form, not inferences from your resume. Blanket automatic rejection based on a resume score is a feature that exists in some configurations and is not how most roles are run. What is far more common is ranking: the recruiter searches or sorts, and works down from the top.

That distinction matters, because it changes what you are optimising for. You are not trying to clear a hidden pass mark. You are trying not to be on page four of a list a human is reading from the top.

The “75% match score” is not a real thing

Different tools compute different numbers by different methods, and none of them is the ATS score, because there is no such number. Any tool promising to tell you “your ATS score” as though it were a property of your resume is selling you a metric it invented. A coverage number can still be genuinely useful — it just has to be honest about being a count of specific things rather than a verdict.

White text and keyword stuffing do not work

Extracted text has no colour. Invisible keywords are perfectly visible to the system and, when a human opens the file, obvious and disqualifying. This trick has been circulating for fifteen years and has never been a good idea.

What actually costs you interviews

Here is the part that is real. Text extraction is imperfect, and when it fails, it fails silently. Your resume is in the system, and what the system has is not what you wrote.

• Text inside images. A resume exported as a picture, or a header graphic containing your name and contact details, extracts as nothing at all.
• Tables used for layout. Extraction walks cells in an order that frequently is not reading order, so dates detach from the jobs they belong to and skills merge into sentences.
• Multi-column layouts. Same problem. Some parsers read across the page rather than down each column, interleaving two unrelated sections into nonsense.
• Headers and footers. Some parsers discard them. Contact details placed there can disappear entirely — you become unreachable, in a system whose purpose is to reach you.
• Unusual section headings. “Where I’ve Been” instead of “Experience”. A human enjoys it; a parser looking for known headings does not find your work history.

None of this is a robot judging you. It is a file conversion going wrong, and it is completely fixable — which is why a template that parses cleanly matters more than any amount of keyword tuning.

The second real problem: vocabulary

Once your text extracts correctly, the remaining question is whether a recruiter searching for the six things they were briefed on will find them in your document. If you ran the incident rotation and the posting says “on-call ownership”, a search for the second does not surface the first. Nothing unfair has happened; the words simply differ.

That is what tailoring fixes, and it is the honest version of “use keywords”: not inventing experience, but describing real experience in the language the reader is looking for. The method is here.

A short, boring checklist

1. Submit a PDF unless the form asks for something else, and make sure it was exported from a document rather than scanned.
2. One column. Keep contact details in the body, not in a header or footer.
3. Conventional section names: Summary, Experience, Skills, Education.
4. No text inside images, and no tables used for layout.
5. Standard dates — “Mar 2022 – Present” parses; “3/22–now” often does not.
6. Fill in the application form fields properly. Knockout questions are usually there, not in your resume.
7. Check that the terms the posting repeats actually appear in your document.

That is the whole thing. It is much less dramatic than the folklore, and it is the part you can actually control.